[BUG] Claude Code Agent Going Offline During Active Development

Resolved 💬 39 comments Opened Jul 9, 2025 by pm0code Closed Jul 27, 2025
💡 Likely answer: A maintainer (dicksontsai, collaborator) responded on this thread — see the highlighted reply below.

Environment:

  • Claude Code Version: Latest (as of 2025-01-09)
  • Model: Opus 4
  • OS: Windows 11 via WSL2

Description:
The Claude Code agent appears to "go offline" or become unresponsive during active development work. The agent is actively working and producing output, but from the user's perspective, it seems to disconnect or pause unexpectedly.

Steps to Reproduce:

  1. Start a complex development task (e.g., implementing canonical framework)
  2. Agent begins research and implementation
  3. Agent creates multiple files and documentation
  4. User perceives agent as "going offline" despite active work continuing

Expected Behavior:

  • Continuous, visible progress updates
  • Clear status indicators when processing
  • No perception of disconnection during active work

Actual Behavior:

  • Agent appears to disconnect or pause
  • User has to prompt "why do you go offline?"
  • Work is actually continuing but not visibly apparent to user

Impact:

  • Disrupts development flow
  • Creates uncertainty about agent status
  • Requires manual intervention to confirm agent is working

Possible Causes:

  1. Long-running file operations without status updates
  2. Complex research/thinking processes without visible output
  3. UI not updating during intensive operations
  4. Missing progress indicators during multi-step tasks

Suggested Fix:

  • Add periodic status updates during long operations
  • Implement progress indicators for multi-step tasks
  • Provide heartbeat/alive indicators during processing
  • Better communicate when agent is thinking vs. offline

Reproducibility: Consistent during complex, multi-file operations

Severity: High - Impacts developer productivity and confidence

Additional Context:
This issue occurred multiple times during canonical framework implementation, causing the user to interrupt work and question agent availability.

View original on GitHub ↗

39 Comments

pm0code · 1 year ago

<img width="1760" height="448" alt="Image" src="https://github.com/user-attachments/assets/a44d030e-882f-4137-81a6-c1cb9d699476" />

pm0code · 1 year ago

<img width="1760" height="419" alt="Image" src="https://github.com/user-attachments/assets/b3f6fb6d-9668-44aa-b06c-596fc46b28d6" />

dicksontsai collaborator · 1 year ago

The 529 error is unrelated (Opus is down 😦 ). Let's focus on the false "offline" display for this issue?

halandi · 1 year ago

I can confirm, i am facing the same issue and causing some MCP disruptions due to this.

Claude Code
Plan: MAX (Opus 4)
Machine: M3 Max (Apple Silicon)
OS: Sequoia (WARP Terminal)

Hopefully we can fix this issue in a reasonable timeline 🤞

TheRealJimShady · 1 year ago

I'm having exactly the same issue with it hanging in 'offline'.

Environment

  • Claude Code: v1.0.45
  • Plan: Max
  • Chip: Apple M1 Pro
  • Memory: 32 GB
  • OS: Sonoma
dicksontsai collaborator · 1 year ago

@halandi I double-checked our client code and I see that the "offline" label only provides visual UI feedback. It does not impact any functionality at all. If you see "disruption", you most likely have network issues on your end.

The main issue we're investigating here is when the "offline" label incorrectly triggers when the user is actually online, causing the user to be confused.

pcj · 1 year ago

Same issue. No other networking issues other than claude:

<img width="1337" height="263" alt="Image" src="https://github.com/user-attachments/assets/eb2c3754-0aa3-492b-ab48-87bbe1d91bd2" />

TheoBong · 1 year ago

Same issue, no networking issues and the terminal becomes stuck (no ability to type, ctrl+c, nothing) for about 5 minutes then comes back and all keystrokes get put in all at once... very weird

hamhanry · 1 year ago

Same issue here. it was intermittent issue, if you wait for some minutes, sometimes it works again. but when generating toker somehow feels not right because there is a pause in generating compared my previous experience using it.

<img width="458" height="42" alt="Image" src="https://github.com/user-attachments/assets/07ecb757-62a6-4be4-9475-4e2f610433b4" />

searchingtofind · 1 year ago

Exactly the same issue. Offline keeps coming on, unresponsive terminal, and limited Opus 4 usage on a Max plan.. (small sessions, 30 miniatures before sonnet tier downgrade)

TheRealJimShady · 1 year ago

Telling claude to break down the work into smaller tasks so that it doesn't hang seems to help.

Astrotope · 1 year ago

Same issue here, Claude Code says it is offline, and the terminal freezes, you cannot "esc" to stop an operation, or enter text in the terminal. It is just frozen. You just have to wait. And this is killing a lot of time. Claude Code is really quite amazing, but this is a showstopper. I'm actually trying out other options, simply out of frustration.

I also see the terminal just freeze up, without the offline message. You cannot "esc" to pause an operation that is running, you just have to wait for it to complete.

I would say the not being able to "esc" from an in-progress task and the appearance of the offline message, are happening around 80-90% of the time at the moment. This wastes a large number of tokens on tasks that would have been terminated, and significantly slows down coding sessions. This is an amazing product, but it is quickly becoming unusable, as it is slowing me down, rather than speeding up what I do.

I am also seeing more of the API timeout errors, where Claude Code, just seems to be waiting for a long time, and then it shows 1/10 timeout errors, and continues 2/10 etc... for a very long time, but with these I can at least exit Claude Code, by typing in the terminal, and restart with claude --continue. Still not great, and you do not know what is going on until the first API timeout error is displayed, which can be a minute or more.

<img width="450" height="22" alt="Image" src="https://github.com/user-attachments/assets/5741f31f-b57e-4c39-9233-b08a6b10cdd0" />

It says 71s, but it was more like 3 minutes, by the time it completed, it just stopped updating the time. This is typical of my experience at the moment.

matthewadams · 1 year ago

Similar. I'm on Linux Mint, terminal with claude code just hung for minutes stating offline, then came back. Curiously, it happened right after I saw Claude Opus 4 limit reached, now using Sonnet 4. I'm on the Max plan.

necipsunmaz · 1 year ago

I'm having exactly the same issue and no networking issues with it hanging in 'offline'.

Environment

Claude Code: v1.0.45
Plan: Max
Chip: Apple M4 Pro
Memory: 64 GB
OS: 15.5 (24F74)

charlesvestal · 1 year ago

Same here, using pro, max, sonnet, or opus.

davidleitch1 · 1 year ago

I have exactly the same issue. Also M2 chip, 64 GB ram, max plan. I wonder if Claude can keep up with demand.

samuelho-dev · 1 year ago

Facing this same issue. m1 64 ram

ahmad-farras · 1 year ago

have the same issue here. bump this up

Samxx97 · 1 year ago

Use claude doctor seems to have fixed the issue for me

⎿ API Error (Request timed out.) · Retrying in 1 seconds… (attempt 1/10)
⎿ API Error (Request timed out.) · Retrying in 1 seconds… (attempt 2/10)
⎿ API Error (Request timed out.) · Retrying in 2 seconds… (attempt 3/10)
⎿ API Error (Request timed out.) · Retrying in 4 seconds… (attempt 4/10)
⎿ API Error (Request timed out.) · Retrying in 8 seconds… (attempt 5/10)
⎿ API Error (Request timed out.) · Retrying in 19 seconds… (attempt 6/10)
⎿ API Error (Request timed out.) · Retrying in 35 seconds… (attempt 7/10)
⎿ API Error (Request timed out.) · Retrying in 35 seconds… (attempt 8/10)

try again

⎿ API Error (Request timed out.) · Retrying in 1 seconds… (attempt 1/10)
⎿ API Error (Request timed out.) · Retrying in 1 seconds… (attempt 2/10)
⎿ API Error (Request timed out.) · Retrying in 33 seconds… (attempt 9/10)
⎿ API Error (Request timed out.) · Retrying in 2 seconds… (attempt 3/10)
⎿ API Error (Request timed out.) · Retrying in 4 seconds… (attempt 4/10)
⎿ API Error (Request timed out.) · Retrying in 8 seconds… (attempt 5/10)
⎿ API Error (Request timed out.) · Retrying in 36 seconds… (attempt 10/10)
⎿ API Error (Request timed out.) · Retrying in 19 seconds… (attempt 6/10)
⎿ API Error (Request timed out.) · Retrying in 38 seconds… (attempt 7/10)
⎿  API Error: Request timed out.

claude doctor
cbaugus · 1 year ago

I have been getting the same issues here.

Claude Code: v1.0.51
Plan: Max
Chip: Apple intel i9
Memory: 32 GB
OS: 15.5

monkCommits · 1 year ago

Exact same issue.
Try switching the internet

AndyWendt · 1 year ago

Having the same issue. I was using it in the evening and it was basically unusable.

Claude doctor didn't seem to do anything. Ubuntu 24.04

AndyWendt · 1 year ago

Looks like for me the issue was that Claude duplicated my hooks... for thousands of lines. It must have been just sitting there running my hooks. lol

dicksontsai collaborator · 1 year ago

@AndyWendt unfortunately, there was a regression in settings that duplicated hooks configuration #3477 . The fix is going out with v1.0.53.

matthewadams · 1 year ago

@AndyWendt Is that fix expected to address all of our problems, or just the duplicate hook invocations?

AndyWendt · 1 year ago
@AndyWendt Is that fix expected to address all of our problems, or just the duplicate hook invocations?

@matthewadams It fixed my issue where it was going offline. But I can't say for other people.

myshadowww · 12 months ago

If it helps others, my issue was because Claude code is trying to tunnel data over UDP:53 and on my network all DNS traffic is redirected via the gateway.

RAJI-Zakaria · 12 months ago

i've run into the same error and i was sure that i've had enough tokens remaining but the error persisted after many attempts.

Solution : Restarting my mac solved the issue.

asolopovas · 12 months ago

yesterday 529 problems today similar issue again...

streamneil · 11 months ago

yes, reboot mac not work. (✻ Ruminating… (6s · ⚒ 230 tokens · esc to interrupt · offline))

dare2live · 11 months ago

Same here, I even completely reinstalled Claude Code TWICE!

avondalepark · 11 months ago

I hate to say 'me too' but it started happening to me a few hours ago - I'm on the $200 plan and have plenty of credit left.

⎿ API Error (Request timed out.) · Retrying in 9 seconds… (attempt 5/10)
⎿ API Error (Request timed out.) · Retrying in 1 seconds… (attempt 1/10)
⎿ API Error (Request timed out.) · Retrying in 19 seconds… (attempt 6/10)
⎿ API Error (Request timed out.) · Retrying in 1 seconds… (attempt 2/10)
⎿ API Error (Request timed out.) · Retrying in 2 seconds… (attempt 3/10)
⎿ API Error (Request timed out.) · Retrying in 38 seconds… (attempt 7/10)

✻ Crunching… (21s · ↑ 0 tokens · esc to interrupt · offline)

eric8810 · 11 months ago

<img width="1376" height="348" alt="Image" src="https://github.com/user-attachments/assets/029b00b7-e78c-4c0b-bf8e-8f0f444bc132" />

I have the same problem, it is working with status shows offline

avondalepark · 11 months ago

Every day this week it goes offline for hours. I've complained to support but they arent helpful. It's time to drop the subscription and try some other model. Claude Code might be the best but it isn't reliable enough anymore.

Zate · 11 months ago

Seeing this too, and if I am running multiple sessions in multiple screens, either in VSCode or via shell to WSL, then I see it in both (or all) of the sessions. It can also be happening when I go to bed, and when I retry in the morning, still happening.

I've tried running with claude -d and ANTHROPIC_LOG=debug and it doesn't show me anything other than the calls timeout. It feels like its external to the client, or its subtle enough that the client isn't seeing it as an error, just as HTTP calls that timeout.

avondalepark · 11 months ago

Still happening today but not as bad probably because of the weekend. I
presume it'll be really bad again tomorrow but I'm working with
gemini-2.5-pro and while it isn't as good at coding, it is much better at
remembering what I want and how I want it built. I know this is weird but
I think it has a nicer 'personality' too - or maybe I just need to touch
grass for a while.

On Sun, Jul 27, 2025 at 1:32 AM Zate @.***> wrote:

Zate left a comment (anthropics/claude-code#3181) <https://github.com/anthropics/claude-code/issues/3181#issuecomment-3123640050> Seeing this too, and if I am running multiple sessions in multiple screens, either in VSCode or via shell to WSL, then I see it in both (or all) of the sessions. It can also be happening when I go to bed, and when I retry in the morning, still happening. I've tried running with claude -d and ANTHROPIC_LOG=debug and it doesn't show me anything other than the calls timeout. It feels like its external to the client, or its subtle enough that the client isn't seeing it as an error, just as HTTP calls that timeout. — Reply to this email directly, view it on GitHub <https://github.com/anthropics/claude-code/issues/3181#issuecomment-3123640050>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABILB4G2B3WPHIPEGN4BU3D3KQMYDAVCNFSM6AAAAACBCTG4WWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTCMRTGY2DAMBVGA> . You are receiving this because you commented.Message ID: @.***>
avondalepark · 11 months ago

Closing it as #3998 is not a good idea. The issue described in that ticket talks about the functionality still working even if the "offline" status is showing. What we have is a complete failure of Claude. Mine goes offline for hours at a time and it is definitely a Claude Code issue as in the window beside it I'm running Gemini and listening to YouTube in the background. Even the claude.ai site is working at the time so it's specifically a claude code problem and it is a total loss of service.

pm0code · 11 months ago
@halandi I double-checked our client code and I see that the "offline" label only provides visual UI feedback. It does not impact any functionality at all. If you see "disruption", you most likely have network issues on your end. The main issue we're investigating here is when the "offline" label incorrectly triggers when the user is actually online, causing the user to be confused.

@dicksontsai, you did not tag me so I missed this.
FYI, network is out of the question. the machine that I am working on is on a 10G internal network with a solid Gig link to internet. I can do anything on this network without any issues. see the shot.

<img width="621" height="900" alt="Image" src="https://github.com/user-attachments/assets/70cc518a-59bf-485d-a4dc-ed50154fbd30" />

github-actions[bot] · 11 months ago

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.